Load Regulates Bone Formation and Sclerostin Expression through a TGFβ-Dependent Mechanism

Bone continually adapts to meet changing physical and biological demands. Osteoblasts, osteoclasts, and osteocytes cooperate to integrate these physical and biochemical cues to maintain bone homeostasis. Although TGFβ acts on all three of these cell types to maintain bone homeostasis, the extent to which it participates in the adaptation of bone to mechanical load is unknown. Here, we investigated the role of the TGFβ pathway in load-induced bone formation and the regulation of Sclerostin, a mechanosensitive antagonist of bone anabolism. We found that mechanical load rapidly represses the net activity of the TGFβ pathway in osteocytes, resulting in reduced phosphorylation and activity of key downstream effectors, Smad2 and Smad3. Loss of TGFβ sensitivity compromises the anabolic response of bone to mechanical load, demonstrating that the mechanosensitive regulation of TGFβ signaling is essential for load-induced bone formation. Furthermore, sensitivity to TGFβ is required for the mechanosensitive regulation of Sclerostin, which is induced by TGFβ in a Smad3-dependent manner. Together, our results show that physical cues maintain bone homeostasis through the TGFβ pathway to regulate Sclerostin expression and the deposition of new bone.


Introduction
Osteocytes coordinate the adaptation of bone to changing physical demands on the skeleton [1]. Upon sensing mechanical load through their canalicular processes, osteocytes initiate a series of biochemical signaling events that coordinate the activity of osteoblasts and osteoclasts to increase bone mass [2]. In this way, physical stimuli employ established biochemical pathways long known to participate in the maintenance of bone homeostasis, including parathyroid hormone (PTH) [3], insulin-like growth factor-I (IGF-1) [4], and prostaglandin signaling (PGE 2 ) [5]. Despite recent progress in deciphering the molecular mechanisms by which physical signals regulate bone homeostasis, many questions remain.
Sclerostin, a secreted protein expressed by osteocytes, responds to mechanical load and antagonizes bone formation [6]. Loss of function mutations in either the sclerostin-encoding gene, SOST, or in its regulatory sequence cause the human syndromes known as sclerosteosis and van Buchem disease, both of which are characterized by high bone mass [7]. Sclerostin acts by binding to the Wnt co-receptor Lrp5/6 on osteoblasts to inhibit Wntinducible osteogenesis [8]. Sclerostin plays a central role in the anabolic response of bone to mechanical loading. Applied mechanical loads repress Sclerostin mRNA and protein expression [6], thereby releasing the brakes on new bone synthesis.
Conversely, Sclerostin-antagonizing antibodies prevent bone loss due to unloading of the bone [9]. Several pathways that control bone development and metabolism also regulate SOST expression. The BMP pathway induces SOST expression during bone development [10]. PTH, an essential regulator of mineral homeostasis, represses SOST expression [11]. This PTH-mediated repression of SOST requires MEF2 recruitment to a highly conserved regulatory region 35-kb downstream from the SOST gene [12]. The rapid increase in PGE 2 following mechanical load also contributes to the mechanosensitive repression of SOST, though the mechanism remains to be identified [13]. The master osteoblast transcription factor Runx2 binds and induces transcription from a more proximal element of the SOST promoter [14]. Although many pathways modulate SOST expression, these pathways do not fully explain the complexity of mechanotransduction in bone and the regulation of SOST expression.
TGFb is a critical regulator of bone homeostasis. Through its effects on osteoblast and osteoclast migration, proliferation, differentiation and viability, TGFb couples bone formation with bone resorption [15,16]. In this way, TGFb maintains both bone mass and bone quality [17][18]. Activated TGFb binds to its receptors, TbRI and TbRII, causing their heterotetramerization and transphosphorylation. Many intracellular proteins, including Smad2, Smad3 and other non-canonical effectors, are phosphor-ylated upon recruitment to the activated TbRI/TbRII complex [19]. Phosphorylated Smad3 translocates to the nucleus where it can activate or repress the activity of sequence-specific transcription factors, such as Runx2 [20]. Through crosstalk at each the ligand, receptor and effector levels, the TGFb pathway integrates signals from multiple stimuli. For example, fluid flow induces TGFb1 mRNA expression in SAOS-2 cells [21]. Cell-generated tension by myofibroblasts converts the TGFb ligand from the latent to active form [22]. PTH receptors drive clustering and internalization of TbRI and TbRII, leading to desensitization of the pathway [23]. The nuclear localization of Smads is sensitive to other biochemical and physical cues, including Wnt signaling [24], cytoskeletal tension [25], and extracellular matrix stiffness [26]. However, the net effect of mechanical load on the overall activity of the TGFb pathway remains unknown. Therefore, we hypothesize that the TGFb pathway integrates signals provided by mechanical load to maintain bone homeostasis, in part by regulating the expression of Sclerostin.
Using a combination of genetically modified mouse models and in vitro approaches, we investigated the effect of mechanical load on TGFb activity, the role of TGFb in load-induced bone formation, and the regulatory relationship between TGFb and Sclerostin. Taken together, our results suggest that TGFb plays a critical role in the mechanosensitive regulation of Sclerostin and is required for the anabolic response of bone to mechanical load.

Mouse models
Treatments and protocols used for the animal studies were approved by the University of California, San Francisco Institution Animal Care and Use Committee (Protocol # AN082159-03A) and were designed to minimize discomfort to the animals. This study used [8][9] week-old male SBE-luciferase mice [27] or DNTbRII mice, which express a dominant negative version of TbRII under control of 1.8 kb of the osteocalcin promoter [28]. Wild type littermates were used as comparative controls.
In vivo mechanical loading of the tibiae Axial compressive loads equivalent to 10 times the mouse's body weight were delivered by a Bose Electroforce ELF3200 desktop load frame (Bose, MN, USA) fitted with two custom-made hemi-spherical fixtures that gripped the mouse knee and ankle. Similar methods of in vivo loading have been shown to upregulate bone anabolism [29]. In our preliminary studies of ex vivo limb loading using in situ strain rosettes, these loading parameters produce maximum principal strains in the range of 1500-2500 me on the mid-diaphyseal surface of the tibiae. For each mouse, only the right hind limb was loaded, while the left hind limb was not loaded to serve as the contralateral control (nonloaded). Each round of loading consisted of 600 cycles of axial compression at 1 Hz administered under general injectable anesthesia. Mice were subjected to this loading regimen once daily for either 1 day (short-term) or once a day for 5 days (prolonged).

In vivo luciferase imaging
Five hours after a single bout of loading, the SBE-luc mice (n = 6) were anesthetized using isoflurane, injected with 150 mg Dluciferin (Xenogen) per kg body weight, and imaged using the IVIS-200 Bioluminescence system 10 minutes after injection (Caliper Bioscience, MA, USA) [22,30]. Photons emitted from living mice were acquired as photons per second per cm 2 per steradian (sr) by using LIVINGIMAGE software (Xenogen) and integrated over 20 min. For photon quantification, a region of interest was manually selected and kept constant within all experiments; the signal intensity was converted into photons/s/ cm 2 /sr. The resulting quantitative measures were segmented and contoured at the tibiae to determine the relative levels of Smad2/3 activity between the loaded and non-loaded limb.

Immunohistochemistry
The loaded and nonloaded tibiae from SBE-luciferase (n = 5), DNTbRII (n = 5), or WT mice (n = 5) were dissected and fixed in 2% paraformaldehyde for 24 h at 4uC. The bones were decalcified in 19% EDTA solution for 7-10 days and decalcification was confirmed with x-ray imaging. Bones were then infiltrated with a series of 10%, 20%, and 30% sucrose solutions [31]. After sucrose infiltration, each tibia was cut into three fragments, with the distal fragment being 8 mm long and the middle fragment being 6 mm long, to ensure that comparable sections of each region were analyzed for each bone. Each fragment was embedded in OCT for frozen sectioning (10 mm sections). The tissue sections were permeabilized with 0.3% Triton X-100, processed for antigen retrieval with Ficin (Invitrogen 00-3007), and blocked for intrinsic peroxidase activity with 3% hydrogen peroxide. For detection of luciferase, the sections were blocked with 1.5% normal goat serum (Vectastain) and incubated with anti-Luciferase primary antibodies (Abcam ab21176) at a dilution of 1:100. For detection of Sclerostin, the sections were blocked in 1.5% normal rabbit serum (Vectastain) and incubated with anti-Sclerostin primary antibodies (R&D Systems AF1589) at a dilution of 1:13 with 0.05% Tween-20. Normal rabbit IgG (Caltag Lab 10500) and normal goat IgG (Santa Cruz sc-2028) were used at the same concentrations as primary luciferase and Sclerostin antibodies, respectively, to control for the specificity of immunostaining. The binding of peroxidase-conjugated secondary antibodies was detected with a DAB kit (Vector Lab).
Three to five immunostained sections for each bone were analyzed quantitatively. A composite of 20X images was generated for each immunostained section, from which all DAB-stained (brown) and unstained lacunae were counted and recorded using ImageJ. The percent of osteocytes expressing luciferase or Sclerostin was determined by dividing the number of DABpositive lacunae by the total number of lacunae for each section. The percent change in luciferase or Sclerostin expression due to load was determined by subtracting the averaged percentage of luciferase or Sclerostin expression in the nonloaded tibia from the loaded tibia as describe [6]. Since quantitative analyses consider the effect of loading only on the number of stained and unstained lacunae, not on the staining intensity, they likely underrepresent the effects of mechanical loading on Sclerostin and luciferase expression.

Micro-computed tomography
Five hours after the last session of in vivo loading of the DNTbRII mice and their WT littermates (n = 6), tibiae were dissected from euthanized mice, fixed in 4% paraformaldehyde in PBS, and serially dehydrated in graded ethanol. The bones were then scanned using micro-CT to determine the relative changes in bone geometry (VivaCT40, Scanco Medical AG). The micro-CT scanner was operated at the peak energy of 70 kVp, current of 114 mA, integration time of 381 ms, and a 10 mm voxel resolution. The scans were segmented using an attenuation constant of 200, and then the structural parameters of bone were evaluated. The proximal tibial trabecular bone was evaluated for changes in trabecular connectivity, trabecular thickness, and volume fraction. The cortical bone changes were evaluated at the tibio-fibular junction for cortical thickness, cross-sectional area, and moment of inertia.

Dynamic histomorphometry
Two intraperitoneal injections of calcein were administered at 0.02 mg/g body weight to DNTbRII mice and their WT littermates (n = 3) that underwent a prolonged 5-day loading regimen. The first injection was administered on the same day as the first loading bout, and the second injection was administered on the same day as the fifth and final loading bout (4 days apart). The mice were euthanized two days after the final loading bout and then the tibiae were collected and fixed in 4% paraformaldehyde in PBS, serially dehydrated in graded ethanol, and embedded in a plastic resin (TechnoVit 5, EMS, Pennsylvania). The embedded blocks were sectioned using a tungsten carbide blade and mounted on glass slides. The mineral apposition rate (MAR, mean inter-label thickness divided by the time between the two labeling periods) was computed at both the periosteal and the endosteal surfaces using ImageJ by evaluating fluorescent micrographs taken with a 20X optical objective. The Mineral Apposition Rate (MAR) was calculated in accordance with the ASBMR nomenclature [32].

Western Analysis
For Western analysis of cortical bone protein, both tibiae were collected 3-5 h after one application of unilateral load (n = 4). Bones were trimmed to remove the distal and proximal epiphyses and flushed with PBS to remove bone marrow. The remaining cortical bone was homogenized with a rotor-stator homogenizer (Omni) in radioimmunoprecipitation assay (RIPA) buffer (10 mM Tris-HCl pH 8, 140 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1% Triton X-100, 0.1% Sodium deoxycholate, 0.1% SDS) supplemented with 5 mM Na 3 VO 4 , 10 mM NaPP i , 100 mM NaF, 500 mM PMSF, and 5 mg/ml eComplete Mini protease inhibitor tablet (Roche) [33]. Western analyses was also used to examinee lysates from SAOS-2 or UMR-106 cells collected after the indicated treatments. Cells were lysed using RIPA buffer on ice. Equal concentrations of the clarified protein from bone or cell lysate were resolved on a 10% SDS-PAGE gel and transferred to a nitrocellulose membrane. Subsequently, the membrane was blocked with 5% milk and the proteins were detected using primary antibodies against pSmad3 (gift from Dr. E. Leof), Smad3 (abcam ab28379), Sclerostin (Santa Cruz, S-19) and b-actin (Abcam, ab8226) and secondary antibodies tagged with an infrared fluorophore. Blots were imaged using a Licor infrared imaging system and are representative of at least 3 experiments.

Statistical Analyses
Statistics were performed using GraphPad Prism 5. T-tests were used to compare the differences between groups of normally distributed data. The Mann-Whitney nonparametric tests were used for non-normally distributed data. Significance of comparisons is defined by p-values equal to or less than 0.05.

Load represses TGFb signaling through Smad2/3
We used the murine one-limb loading model [29] to evaluate the role of TGFb in load-induced bone formation. Mechanical stimulation was applied to one tibia of anesthetized SBE-Luc transgenic mice, which express luciferase under control of a Smad2/3-responsive synthetic promoter [30] [34]. By assessing the functional outcome of Smad2/3 activity, this approach evaluates the net effect of mechanical load on canonical TGFb signaling -whether those effects occur at the ligand, receptor, or effector levels. Five hours after loading, in vivo bioluminescent imaging revealed that Smad2/3 reporter activity was consistently reduced in the loaded limb relative to the nonloaded limb of the same mouse ( Figure 1A). Loading significantly repressed TGFbmediated Smad2/3 reporter activity by an average of 40% ( Figure 1B). Immunolocalization of luciferase expression confirms that the load-mediated repression of Smad2/3 activity occurs in osteocytes within the cortical bone. In addition to generalized reduction in luciferase staining of cortical bone ( Figure 1C), the number of luciferase positive osteocyte lacunae was significantly reduced in response to mechanical loading ( Figure 1D). To further examine the mechanism by which load represses Smad2/3mediated transactivation, we compared the effect of in vivo mechanical loading to the effect of in vitro TGFb stimulation or inhibition on Smad3 phosphorylation. The level of phosphorylated Smad3 in the loaded cortical bone was consistently reduced relative to that in the nonloaded bone of the same animal ( Figure 1E, lower panel). The magnitude of this load-dependent  is comparable to that achieved in UMR-106 cells treated with an inhibitor of the TGFb type I receptor ( Figure 1E, upper  panel). Therefore, the phosphorylation and activity of the key TGFb effector, Smad3, is rapidly and significantly reduced in osteocytes following mechanical loading of the limb.
The anabolic response of bone to mechanical load requires TGFb signaling We next examined whether the mechanosensitive regulation of TGFb is necessary for load-induced bone formation. Mechanical load was applied to one limb of DNTbRII mice that have impaired TGFb signaling due to expression of a dominant negative TGFb type II receptor under control of the osteocalcin promoter [28]. In wild-type littermates, 5 days of mechanical loading stimulates a 19% increase in trabecular bone volume fraction (BV/TV) relative to the nonloaded limb (Table 1), similar to other studies [35,36]. However, mechanical stimulation in DNTbRII mice only increases bone formation by 9%. Mechanical load also failed to stimulate the same magnitude of increase in DNTbRII trabecular connectivity (Tb. Conn), cortical thickness (Cort. Th.), and moment of inertia (MOI) compared to wild-type animals ( Figure 2A, Table 1). Impaired TGFb signaling in DNTbRII mice compromises load-induced bone formation at both the periosteal and endosteal surfaces, as determined by the reduced fluorochrome intensity in the loaded DNTbRII bone compared to the loaded WT bone ( Figure 2B). Despite a slightly increased basal mineral apposition rate (MAR) in DNTbRII mice, relative to WT, mechanical load was unable to further stimulate a significant increase in the DNTbRII MAR. In contrast, mechanical load stimulated large (.40%) and significant increases in WT MAR relative to the nonloaded WT limb ( Figure 2C). These results demonstrate that the anabolic effect of mechanical load on bone formation requires an active TbRII-dependent signaling pathway.

TGFb sensitivity is required for regulation of Sclerostin by mechanical load
Since the repression of SOST/Sclerostin by mechanical load is a key event in bone anabolism [6,37], we sought to determine if SOST regulation requires an intact TGFb signaling pathway. We evaluated the effect of mechanical load on Sclerostin expression in DNTbRII mice. Following 5 days of loading, Sclerostin expression in wild-type cortical bone is reduced ( Figure 3A). Specifically, the percentage of Sclerostin positive lacunae decreases by 9% following mechanical loading of the tibia ( Figure 3B). However, in DNTbRII bone, the application of mechanical load produced no distinguishable difference in Sclerostin expression between the loaded and control limbs ( Figures 3A, 3B), suggesting that the load-mediated regulation of Sclerostin requires sensitivity to TGFb.

TGFb signaling through Smad3 induces SOST expression
To examine mechanisms by which TGFb sensitivity is required to regulate SOST, we first evaluated the effect of TGFb on SOST mRNA levels in UMR-106 osteosarcoma cells. TGFb rapidly induces SOST mRNA expression, with maximal induction following 8 h of treatment ( Figure 4A). Conversely, SOST mRNA expression is repressed by a specific inhibitor of the TGFb type I phosphorylated Smad3, total Smad3, and b-actin by Western analysis. UMR-106 cells were harvested 2 h after treatment with vehicle (DMSO), TGFb1 (5 ng/ml), or an inhibitor of the TGFb type 1 receptor, Alk5 (TbRI-I, SB431542). Tibiae were harvested 3h after loading (n = 4 mice). Loaded tibiae (L) were compared to the nonloaded tibiae (NL) from the same mouse. (* p,0.05). doi:10.1371/journal.pone.0053813.g001 Since mechanical load represses TGFb signaling through Smad2/3 in osteocytes (Figure 1), as well as SOST expression [6], we hypothesized that TGFb induction of SOST is Smad2/3dependent. Even without exogenous TGFb, cotransfected Smad3 was sufficient to activate a promoter-reporter construct that expresses luciferase under control of the human SOST promoter and 3 copies of a previously identified SOST-regulatory enhancer sequence (ECR5) ( Figure 4B). Cotransfected Smad3 further enhanced the TGFb-inducibility of this construct.

TGFb/Smad3 induction of SOST is indirect and insensitive to Runx2
Though Smad3 increases SOST-reporter activity, these effects are indirect. As expected, incubation of UMR-106 cells with actinomycin D, an inhibitor of transcription, blocks TGFbinducible SOST mRNA expression ( Figure 4C). However, TGFb-inducible SOST mRNA expression is also completely abrogated in the presence of cycloheximide, an inhibitor of translation, even at the earliest 2 h time point ( Figure 4C). These data suggest that TGFb-inducible SOST expression occurs through an indirect Smad3-mediated pathway.
TGFb and Smad3 regulate the expression and activity of the osteoblast transcription factor Runx2 [20]. Since SOST is a Runx2target gene [14], we sought to determine if Runx2 was required for  the indirect TGFb/Smad3-dependent regulation of SOST expression. Runx2-targetting siRNA yielded a 35-80% decrease in Runx2 mRNA levels in UMR-106 and SAOS cells (not shown), with a corresponding reduction in Runx2 protein expression (Figure 4D). This treatment was sufficient to reduce the expression of Runx2-inducible RANKL by 35% (not shown). However, even with reduced levels of Runx2, the TGFb-mediated induction of SOST mRNA was intact ( Figure 4D), demonstrating that TGFbmediated induction of SOST is insensitive to the level of Runx2 activity. Together these findings show that TGFb regulates SOST mRNA expression indirectly through a Smad3-dependent, Runx2-insensitive mechanism.

Discussion
Here, we report that mechanical load represses TGFb activity, which is required for load-induced bone formation and the regulation of Sclerostin, an inhibitor of bone anabolism ( Figure 5). Loading of mice tibiae rapidly inhibits phosphorylation of Smad3, a TGFb effector, and consequently represses Smad3 activity in osteocytes. TGFb signals through Smad3 to indirectly stimulate SOST mRNA expression. Furthermore, intact TGFb signaling is required for load to repress Sclerostin expression and induce bone formation. Taken together, our results demonstrate that TGFb plays a critical role in the mechanosensitive regulation of Sclerostin and is required for the anabolic response of bone to mechanical load.
Several lines of evidence implicate TGFb as a regulator of bone homeostasis. Many mouse models with mutations in components of the TGFb pathway have altered bone mass and bone matrix material properties, resulting from disruption of the tightly controlled balance between osteoblast and osteoclast activity. For example, reducing TGFb signaling in mice by expression of a dominant negative TbRII allele in the osteoblast lineage increased bone mass by indirect reduction in osteoclast activity [28]. In addition, mice deficient in Smad3 have lower bone mass but an increased bone matrix elastic modulus indicating the importance of TGFb in the maintenance of bone mass and bone quality [38]. Furthermore, inhibiting TGFb signaling with a chemical inhibitor of TbRI affects postnatal bone formation and bone quality [30]. Stimuli that shift bone metabolism, including mechanical load, exert their effects by acting on pathways that normally maintain homeostasis. The extent to which TGFb mediates the anabolic effect of mechanical load on bone has not previously been shown. Here, we show that disrupting TGFb sensitivity in osteoblasts prevents mechanical load from inducing bone formation. This suggests that mice with mutations in the TGFb pathway uncouple the normal regulation of bone mass from stimuli such as mechanical load; likely contributing to the high or low bone mass phenotypes observed in a several mouse models and human diseases in which TGFb signaling is deregulated.
Mechanical load regulates TGFb signaling at multiple levels. Cell-generated tensile stress induces the activation of the latent TGFb ligand by myocytes [22]. In vitro fluid flow stimulates the expression of TGFb1 mRNA in osteoblast cells [21]. Also, loading of the rat ulna induces TGFb1 mRNA expression in the periosteal bone within 4h [39]. Still unclear, however, is the net functional effect of these mechanosensitive changes at multiple hierarchical levels of the TGFb pathway. Unexpectedly, our results show that mechanical load rapidly inhibits Smad3 phosphorylation and, thus, represses the activity of key TGFb effectors, Smad2 and Smad3, in osteocytes. Therefore, mechanical stimulation regulates the TGFb pathway at the transcriptional level, as described above, as well as through post-transcriptional regulation of Smad2/3 activity.
TGFb can be added to the list of pathways found to be mechanosensitive in bone that also regulate Sclerostin. PGE 2 regulates both osteoblast and osteoclast activities and is produced within 5 mins of mechanical stimulation [40]. Once mechanically stimulated, PGE 2 signals through the EP4 receptor to repress SOST transcription in osteoblast cells [41]. Several lines of evidence implicate PTH in load regulation of SOST activity. First, PTH levels are elevated in serum during high impact exercises such as running [3]. Second, SOST overexpression desensitizes mice to PTH-induced bone formation, suggesting that SOST acts as a downstream target of PTH [42][43]. Third, PTH represses SOST expression by inhibiting MEF2 transcriptional activity at the SOST enhancer [12] [11]. Some gaps in the connection between load, PGE 2 and PTH, and SOST remains to be filled.
Here we find that load fails to repress Sclerostin expression by osteocytes or new bone formation in the tibia of DNTbRII mice. Along with the mechanosensitive regulation of Smad2/3 function in osteocytes, these findings establish a clear link between mechanical load, TGFb signaling, SOST levels, and new bone formation. It is important to note that modest bone formation was detected in the loaded tibia of DNTbRII mice (3-9%). Clearly, multiple other pathways in addition to TGFb cooperate to regulate bone anabolism following mechanical load. Since the PTH and TGFb pathways induce desensitization of one another via receptor internalization [23], some of the PTH-dependent effects may occur through TGFb-dependent mechanisms. SOST may be a target of crosstalk between the PTH and TGFb pathways. For example, the SOST regulatory transcription factor, MEF2, is repressed by PTH in osteoblasts [12] as well as by TGFb/Smad3 in myocytes [44]. Furthermore, MEF2 is required for the activation of SOST transcription by TGFb in osteogenic cells [45]. Thus, MEF2 may be a common target of both PTH and TGFb and a point of convergence in their regulation of SOST and load-induced bone formation. Although mechanical load represses Smad2/3 activity in osteocytes, it remains possible that load regulation of SOST through TGFb occurs indirectly through the effects of TGFb on osteoblasts since DNTbRII is not expressed exclusively in osteocytes but also in osteoblasts. Further studies using osteocyte-specific mutations would clarify the precise role of TGFb in each cell type in the response to load.
In conclusion, we find that load represses Smad3 activity in osteocytes and signals through the TGFb pathway to maintain bone homeostasis. TGFb indirectly induces SOST expression, which is mediated by Smad3. Although Runx2 was previously shown to regulate SOST expression [14], the TGFb and Smad3dependent induction of SOST is insensitive to the level of Runx2 activity. Finally, our data reveal a novel mechanism in load regulation of bone formation, which could provide insights for treating bone diseases such as osteoporosis.